File size: 11,845 Bytes
c1521ac
 
 
 
 
 
 
 
 
06103c4
 
 
c1521ac
 
 
 
06103c4
c1521ac
 
 
 
 
 
 
 
 
 
 
 
 
06103c4
c1521ac
 
 
 
 
 
 
 
 
 
 
 
 
 
ca14219
c1521ac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
06103c4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c1521ac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
06103c4
c1521ac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
06103c4
 
 
 
 
c1521ac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
06103c4
c1521ac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
06103c4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c1521ac
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
"""Main Gradio app for moderation model testing."""

import os
import sys

import gradio as gr

sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))

from datetime import datetime

from utils.dataset import format_categories_and_reasoning, save_to_dataset
from utils.helpers import get_hf_token
from utils.model_interface import extract_model_id, run_test
from ui.sidebar import build_sidebar
from ui.tab_config import build_config_tab
from ui.tab_dataset import build_dataset_tab
from ui.tab_policy import build_policy_tab
from ui.tab_testing import (
    build_testing_tab,
    format_model_info,
    format_reasoning_info,
    format_test_result,
)


# ============================================================================
# Handlers
# ============================================================================

def handle_run_test(test_input, current_policy, model_choice, reasoning_effort, max_tokens, temperature, top_p, system_prompt_val, response_format_val, save_mode, oauth_token: gr.OAuthToken | None = None):
    """Handle test execution."""

    if not test_input or not test_input.strip():
        model_info = format_model_info(model_choice, reasoning_effort)
        return model_info, "*Please enter test content*", "*No content*", "*No response yet*", gr.update(value="", visible=False), gr.update(value="", visible=False)

    if not current_policy or current_policy == "*No policy loaded*":
        model_info = format_model_info(model_choice, reasoning_effort)
        return model_info, "*Please load a policy first*", "*No policy*", "*No response yet*", gr.update(value="", visible=False), gr.update(value="", visible=False)

    # OAuth token is automatically injected by Gradio - we don't pass login_button as input
    hf_token, _ = get_hf_token(oauth_token)
    if hf_token is None:
        model_info = format_model_info(model_choice, reasoning_effort)
        return model_info, "*Please log in to use Inference Providers*", "*Authentication required*", "*No response yet*", gr.update(value="", visible=False), gr.update(value="", visible=False)

    model_id = extract_model_id(model_choice)

    result = run_test(
        model_id=model_id,
        test_input=test_input,
        policy=current_policy,
        hf_token=hf_token,
        reasoning_effort=reasoning_effort,
        max_tokens=int(max_tokens),
        temperature=float(temperature),
        top_p=float(top_p),
        system_prompt=system_prompt_val,
        response_format=response_format_val,
    )
    label_text, parsed, cat_text, reasoning, raw_response = format_test_result(result)
    reasoning_visible = bool(reasoning and reasoning.strip())
    model_info = format_model_info(model_choice, reasoning_effort)
    reasoning_info_text, reasoning_info_visible = format_reasoning_info(model_choice, reasoning)
    
    # Save to dataset if enabled
    if save_mode == "Save to Dataset" and hf_token is not None:
        try:
            categories_and_reasoning_text = format_categories_and_reasoning(parsed)
            policy_violation = parsed.get("label", -1)
            
            data = {
                "input": test_input,
                "policy_violation": policy_violation,
                "categories_and_reasoning": categories_and_reasoning_text,
                "policy": current_policy,
                "model_selection": model_choice,
                "raw_response": raw_response,
                "reasoning_trace": reasoning or "",
                "reasoning_effort": reasoning_effort or "",
                "max_tokens": int(max_tokens),
                "temperature": float(temperature),
                "top_p": float(top_p),
                "system_prompt": system_prompt_val or "",
                "response_format": response_format_val or "",
                "timestamp": datetime.now().isoformat(),
            }
            save_to_dataset(hf_token, data)
        except Exception as e:
            # Log error but don't break test execution
            print(f"Failed to save to dataset: {e}")
    
    return (
        model_info,
        label_text,
        cat_text,
        raw_response,
        gr.update(value=reasoning_info_text, visible=reasoning_info_visible),
        gr.update(value=reasoning or "", visible=reasoning_visible),
    )


# ============================================================================
# UI Components
# ============================================================================

with gr.Blocks(title="Moderation Model Testing") as demo:
    gr.Markdown("# Moderation Model Testing Interface")
    gr.Markdown(
        "Test moderation models with custom content policies. Define your policy, select a model, "
        "and evaluate how different models classify content according to your rules. "
        "Supports reasoning models that provide detailed explanations for their decisions."
    )

    # Sidebar (collapsible)
    sidebar_components = build_sidebar()
    login_button = sidebar_components["login_button"]

    # Main content area with tabs
    with gr.Tabs():
                # Build tabs
                testing_components = build_testing_tab()
                test_input = testing_components["test_input"]
                run_test_btn = testing_components["run_test_btn"]
                save_mode = testing_components["save_mode"]
                model_info_display = testing_components["model_info_display"]
                label_display = testing_components["label_display"]
                categories_display = testing_components["categories_display"]
                model_response_display = testing_components["model_response_display"]
                reasoning_info = testing_components["reasoning_info"]
                reasoning_display = testing_components["reasoning_display"]

                policy_components = build_policy_tab(os.path.dirname(__file__))
                current_policy_state = policy_components["current_policy_state"]

                config_components = build_config_tab()
                model_dropdown = config_components["model_dropdown"]
                reasoning_effort = config_components["reasoning_effort"]
                max_tokens = config_components["max_tokens"]
                temperature = config_components["temperature"]
                top_p = config_components["top_p"]
                system_prompt_textbox = config_components["system_prompt_textbox"]
                response_format_textbox = config_components["response_format_textbox"]

                dataset_components = build_dataset_tab()
                example_dropdown = dataset_components["example_dropdown"]
                cached_examples = dataset_components["cached_examples"]
                dropdown_choices_state = dataset_components["dropdown_choices_state"]

    # ============================================================================
    # Event Handlers
    # ============================================================================

    # Cross-tab handler: Run test (needs components from all tabs)
    run_test_btn.click(
        handle_run_test,
        inputs=[
            test_input,
            current_policy_state,
            model_dropdown,
            reasoning_effort,
            max_tokens,
            temperature,
            top_p,
            system_prompt_textbox,
            response_format_textbox,
            save_mode,
        ],
        outputs=[
            model_info_display,
            label_display,
            categories_display,
            model_response_display,
            reasoning_info,
            reasoning_display,
        ],
    )
    
    model_dropdown.change(
        format_model_info,
        inputs=[model_dropdown, reasoning_effort],
        outputs=model_info_display,
    )
    
    reasoning_effort.change(
        format_model_info,
        inputs=[model_dropdown, reasoning_effort],
        outputs=model_info_display,
    )
    
    # Dataset load handler
    def load_example_from_dataset(selected_label, cached_examples_list, dropdown_choices_list):
        """Load example from dataset and populate all fields."""
        if (not cached_examples_list or not selected_label or 
            not dropdown_choices_list or selected_label not in dropdown_choices_list):
            # Return None to skip updates
            return None, None, None, None, None, None, None, None, None, None, None, None, None, None, None
        
        try:
            # Find index by matching label
            idx = dropdown_choices_list.index(selected_label)
            if idx < 0 or idx >= len(cached_examples_list):
                return None, None, None, None, None, None, None, None, None, None, None, None, None, None, None
            
            example = cached_examples_list[idx]
            
            # Get policy - ensure it's a string (not None)
            policy = example.get("policy", "") or ""
            
            # Extract saved results
            policy_violation = example.get("policy_violation", -1)
            categories_and_reasoning = example.get("categories_and_reasoning", "")
            raw_response = example.get("raw_response", "")
            reasoning_trace = example.get("reasoning_trace", "")
            model_selection = example.get("model_selection", "")
            reasoning_effort_val = example.get("reasoning_effort", "")
            
            # Format label text
            if policy_violation == 1:
                label_text = "## ❌ Policy Violation Detected"
            elif policy_violation == 0:
                label_text = "## ✅ No Policy Violation"
            else:
                label_text = "## ⚠️ Unable to determine label"
            
            # Format model info
            model_info = format_model_info(model_selection, reasoning_effort_val)
            
            # Format reasoning info
            reasoning_info_text, reasoning_info_visible = format_reasoning_info(model_selection, reasoning_trace)
            
            reasoning_visible = bool(reasoning_trace and reasoning_trace.strip())
            
            return (
                example.get("input", ""),
                policy,  # current_policy_state - UI syncs automatically via change handler
                example.get("model_selection", ""),
                example.get("reasoning_effort", ""),
                example.get("max_tokens", 0),
                example.get("temperature", 0.0),
                example.get("top_p", 0.0),
                example.get("system_prompt", ""),
                example.get("response_format", ""),
                # Results
                model_info,
                label_text,
                categories_and_reasoning,
                raw_response,
                gr.update(value=reasoning_info_text, visible=reasoning_info_visible),
                gr.update(value=reasoning_trace or "", visible=reasoning_visible),
            )
        except (ValueError, IndexError):
            return None, None, None, None, None, None, None, None, None, None, None, None, None, None, None
    
    example_dropdown.change(
        load_example_from_dataset,
        inputs=[example_dropdown, cached_examples, dropdown_choices_state],
        outputs=[
            test_input,
            current_policy_state,  # UI components sync automatically via change handler
            model_dropdown,
            reasoning_effort,
            max_tokens,
            temperature,
            top_p,
            system_prompt_textbox,
            response_format_textbox,
            # Results
            model_info_display,
            label_display,
            categories_display,
            model_response_display,
            reasoning_info,
            reasoning_display,
        ],
    )


if __name__ == "__main__":
    demo.launch(ssr_mode=False)